Work has progressed in family-based statistical methods for studying genetic effects. Case-control studies aimed at elucidating genetic contributors to the etiology of diseases are problematic because of the 'admixture' problem: If a particular variant allele is to be studied, there may be subpopulations that simultaneously have elevated prevalence of the variant allele and increased risk of the defect, for unrelated reasons. Such an admixture can produce biased estimation in a traditional population-based case-control study. A family-based design avoids this problem by effectively conditioning on the parental genotypes. One can study affected individuals and their parents, who together form a 'triad' of genotypes. Using the triads from such a study, under assumed Mendelian inheritance, one can estimate relative risks for an allelic variant and can differentiate effects that depend on the prenatal effects of the maternal genotype from effects mediated by the (correlated) offspring's inherited genotype. Triads that are incomplete because one or both parents are not available for genotyping can be fully used by applying the Expectation-Maximization algorithm. Further work allows for detection of effects that differ according to the parent of origin of the variant allele, a phenomenon known as 'imprinting.' We are now developing SAS software to implement these methods, which we intend to make available on the web. Further work has applied within-cluster paired resampling to sibship data, which can be a powerful design for studying a disease with onset in later life, when parents would not necessarily be available. Standard methods, such as conditional logistic regression, are not valid for sibships because of the complex correlation structures that are induced among siblings by genetic linkage. Repeated paired resampling solves the problem by letting each sibship repeatedly contribute just one case-control (affected-unaffected) pair. One carries out a logistic analysis of the pairs, and then pools the results across many such analyses, to derive an interpretable estimator for the odds ratio together with a valid estimate of its standard error. We showed that this method outperforms existing methods for genetic sibship analysis.
Showing the most recent 10 out of 42 publications
